Active Control for Object Perception and Exploration with a Robotic Hand

Abstract

We present an investigation on active control for intelligent object exploration using touch with a robotic hand. First, uncertainty from the exploration is reduced by a probabilistic method based on the accumulation of evidence through the interaction with an object of interest. Second, an intrinsic motivation approach allows the robot hand to perform intelligent active control of movements to explore interesting locations of the object. Passive and active perception and exploration were implemented in simulated and real environments to compare their benefits in accuracy and reaction time. The validation of the proposed method were performed with an object recognition task, using a robotic platform composed by a three-fingered robotic hand and a robot table. The results demonstrate that our method permits the robotic hand to achieve high accuracy for object recognition with low impact on the reaction time required to perform the task. These benefits make our method suitable for perception and exploration in autonomous robotics.